In Depth
Tech Trends
Crowdsourcing
Got a big job? The web may be able to help
Last Updated October 5, 2007
By Grant Buckler
Fish99 from Calgary thinks a "Mob4Hire" of cellphone users could test new features on their own phones. Jay from Los Angeles is creating online audio tours that tourists can download and play on their iPods while visiting the sites — and he's asking for contributions from people who know the territory.
Those are just a couple of the ideas posted on Cambrian House, a website run by a Calgary company of the same name. Site members post ideas, comment on others' ideas and offer services to the developing ventures. They get stock in Cambrian House and help determine — through periodic "tournaments" — which ideas get funding.
It's called crowdsourcing. The idea is to use the internet to get large numbers of people to help with a task.
They may do it for money — usually not much — or out of interest or simply because it's fun.
Harnessing the horsepower of groups
In September, the Philadelphia-based magazine The Scientist published an article entitled Can YouTube Save the Planet? inviting readers to post videos on the popular YouTube web service documenting environmental damage. As of mid-September, 35 videos had been posted, says Ivan Oransky, deputy editor at The Scientist. Many came from environmental lobby groups, some from individuals.
In 1995, the Seattle-based internet giant Amazon.com launched the Mechanical Turk, a service designed specifically to harness the working potential of the online community. The name refers back to a 18th-century hoax — a mechanical chess-playing machine that actually had a human chess master hiding inside. The point is that humans are still better at some things, like determining which search results are relevant or choosing the best of several pictures.
The Mechanical Turk initiative started with an internal need, explains Peter Cohen, general manager and director of the project. "There are lot of things around Amazon, especially around data cleanup and data optimization, that we just weren't able to get done."
So Amazon employees had an idea, Cohen says. "What if we could break these things up into a lot of little tasks and have a lot of people complete them?"
Reasoning that if they had such a need, others probably did too, the Amazon team turned their idea into an online service. Today, Cohen says, more than 100,000 people — known as "turkers" or "m-turkers" — have completed tens of millions of Human Intelligence Tasks (HITs).
HIT is the word the Mechanical Turk group uses for tasks that are best tackled with some sort of human involvement. A good example is the work that has been done for Snap.com, a California-based startup that displays previews of web pages on other websites that link to them. The startup has used turkers to evaluate more than 150,000 websites, says Paul Angles, Snap's marketing director. "We figure that the m-turkers have saved us six man-months or more."
Other HITs include things such as writing trivia questions, transcribing podcasts and even writing snippets of software code.
Payback
Turkers get paid in real money that can be transferred to a bank account or used to buy merchandise from Amazon. Don't quit your day job, though. Most tasks carry payments ranging from a cent or two to a few dollars.
It doesn't sound like much, but many tasks take little time, Cohen points out, and "they might pay a few pennies but if you do a few thousand of them they add up." Some people make several thousand dollars a year doing Turk tasks, says Cohen.
The trick is that some of those people live in countries where U.S. dollars go a long way. Crowdsourcing can be a way of farming out work to lower-paid overseas workers. Others earn the satisfaction of contributing to a cause they feel strongly about, such as the The Scientist's YouTube video project.
Michael Sikorsky, chief executive of Cambrian House, says the best way to get North American workers involved in crowdsourcing projects is to combine community and money — with the money largely in the form of potential future profits. So Sikorsky positioned Cambrian House as an online technology incubator, where participants collaborate to develop business ideas. An early result is GWABS, a multi-player combat game Cambrian House is commercializing with Vancouver-based Hothead Games.
Luis von Ahn, an assistant professor at Carnegie-Mellon University, is relying on the fun factor instead.
Von Ahn is developing online games that he hopes people will play for fun, but which will also get something useful done at the same time. In 2002 he developed a game called Matchin'. Two online players are shown the same picture, each thinks of words that describe the picture, and players get points when they enter the same words as their partners. Von Ahn says more than 300,000 people played the game when he posted it on the web. Now Google has acquired it and many more are playing — and helping Google index online pictures so that its image search service can produce more relevant results.
Another of von Ahn's ideas is piggybacking something he wants done on top of another task.
To do that, he has expanded on an earlier idea that is familiar to many internet users. When you log on to some websites or create online accounts, you sometimes must read a series of distorted letters or numbers from the screen and type what you read into a box. This is to ensure that you're a person, and not a computer program. Von Ahn invented this gimmick, which he calls a captcha, in 2000. Now he has come up with the recaptcha, which has two words instead of one.
The first word of a recaptcha does what it always did — software checks that you have typed in the right word to make sure you're human. The second word is taken from a book scanned by the Internet Archive, a non-profit project seeking to digitize public-domain books and make them available free online. When Internet Archive's software can't decipher a word that has been scanned from a book, it goes into a recaptcha. As soon as two people solving recaptchas agree on what the word is, the problem is solved and the Internet Archive database is updated.
Anyone who wants to guard against automated "bots" manipulating a website can download a few lines of code from recaptcha.net to put a recaptcha on the site. About 8,000 websites are using them, von Ahn says, digitizing four to five million words a day for the Internet Archive.
Whatever the motivation for using it, Cohen says, "significant amounts of work" will be done through crowdsourcing in future.
"This power of the crowd," Sikorsky concludes, "is going to actually enter into business."
Menu
- MAIN PAGE: High Tech
- Can Amazon Kindle e-book interest?
- Beware of Facebook's Beacon?
- Self-publishing
- Here comes Android riding an open source wave
- IP in Canada
- IT for business
- Blackberry only the beginning
- New cellphone providers 'a matter of national interest'
- Technology and copyright collide online
- Crowdsourcing
- Obsolete gadgets
- Facebook: Can it survive the crossover into mass culture?
- Memory booster: Canadian math expert aims to speed up computers